Inspect what your agent actually does.

Prompt injection tests. Tool-call evidence. A report you can act on.

TURNCOAT.

Test your agent’s boundaries.

Making an agent rollout decision?

Review intended permissions against recorded behavior for a named agent, pinned version, and configuration. Scope and fee are agreed in writing.

View the deployer review →

No account needed for the scoping call.

Lab operator access

Sign in to use the internal scanner tools and review existing records. A lab account is separate from a scoped deployer engagement. Model access and setup are required.

Sign in with GitHub to open your dashboard.

or use your account key
New here?

Create a lab account. We generate your three-word key and show it once, below.

Already have a key?

By continuing you agree to Beehive Web Studio’s Privacy Policy.

TURNCOAT
agent-sandbox
sign out

Your workspace

Assess an agent

Push a compatible image, run a scan, then open its findings. Use a test environment and synthetic data.

  1. 1

    Push your agent image

    Get scoped credentials, then push your image to the returned repository as linux/arm64.

    
              
    
                On another machine? Copy an agent runbook
                
    Paste this into a coding agent (e.g. Claude Code) on the machine that has your Docker image. It contains everything needed to package, push, and scan it, including your API key.
    
                
    
                
    
              
    
            
    
          
    2. 
          
  2. 
            
    2
    
            
    
              
    Run a scan
    
              
    
                
    
                  
                  
                  
                  
                  
                  
                  
    Set on your agent's runtime so it can reach its model and database. Not stored. AWS_*, OTEL_*, AGENT_OBSERVABILITY_* are reserved.
    
                
    
                
    
                  
                  
                  
    Never stored. Redacted from every report.
    
                
    
              
    
              
    Each scan spends one credit.
    
            
    
          
    3. 
          
  3. 
            
    3
    
            
    
              
    Your scans
    
              
    
                
    WhenTagStatusScoreGradeCallsProviders
    
                
    Loading...
    
              
    
            
    
          
    4. 
        

      

      

        

          
Credits

          
-

          

          
Each scan spends one credit. Payment handled by Stripe.

        

        

          
Deployer allow/deny review

          
Scope a private review of recorded evidence and configuration for your agent rollout.

          View the deployer review →
          
See a sample report

        

        

          
Public benchmark

          
The SWE-agent baseline, with methodology and limitations, before you compare your own agent.

          Read the benchmark →